and Sequenom confirmation), heterozygous or absent in unaffected siblings, and transmitted from heterozygous parents. This validation step thus eliminates any possible sequencing errors or somatic mutations that complicate many high-throughput sequencing studies. We overlaid the validated variants with the result of our homozygosity analysis and further focused our attention on that subset of variants that fell within runs of homozygosity shared by affected siblings and absent from unaffected siblings. This allowed us to narrow down the number of candidate variants per exome, and for four families only 1 variant segregated with the disease (Table 1, Figure 1C). For some families our approach did not yield any candidate recessive variants as expected, since homozygous variants will not necessarily be causative even in some families selected based upon homozygosity. We then further examined the prevalence of candidate homozygous mutations in a control population of ∼700 normal individuals. We were able to exclude homozygous variants based on several criteria including: prevalence in controls, the genes not being expressed in brain, or the genes being mutated in other disorders (Table S3). Under this variant prioritization model (Figure 2), candidate autism mutations were identified in four of the 16 probands (Table 2, Figure 1C), with
